Professional Documents
Culture Documents
On The Effectiveness of Feed-In Tariffs in The Development of Photovoltaic Solar
On The Effectiveness of Feed-In Tariffs in The Development of Photovoltaic Solar
Elbert Dijkgraaf1
Tom van Dorp2
Emiel Maasland3
1 Erasmus School of Economics, Erasmus University Rotterdam, and Tinbergen Institute, the
Netherlands;
2 Solarplaza International BV, the Netherlands;
3 Erasmus University Centre for Contract Research and Business Support (ERBS) BV, the Netherlands.
Tinbergen Institute is the graduate school and research institute in economics of Erasmus University
Rotterdam, the University of Amsterdam and VU University Amsterdam.
Duisenberg school of finance is a collaboration of the Dutch financial sector and universities, with the
ambition to support innovative research and offer top quality academic education in core areas of
finance.
Tom P. van Dorpb
Emiel Maaslandc
a
Erasmus School of Economics and Tinbergen Institute, Erasmus University Rotterdam, Burg.
Oudlaan 50, 3062 PA Rotterdam, The Netherlands
b
Solarplaza International BV, Stationsplein 45, 3013 AK Rotterdam, The Netherlands
c
Erasmus University Centre for Contract Research and Business Support (ERBS) BV, Erasmus
University Rotterdam, Burg. Oudlaan 50, 3062 PA Rotterdam, The Netherlands
ABSTRACT
Growing concern for climate change and rising scarcity of fossil fuels prompted governments
to stimulate the development of renewables. This paper empirically tests whether feed‐in
tariff (FIT) policies have been effective in the development of photovoltaic solar (PV), explicitly
taking into account structure and consistency of FITs. Panel data estimations are employed
for 30 OECD member countries over the period 1990‐2011. We find a positive effect of the
presence of a FIT and the development of a country’s share of PV in the electricity‐mix. This
effect increases if policies are consistent. Tariff height is the most important characteristic of
a FIT, but other characteristics such as cost level, duration of contract and restrictions on
capacity levels can also not be neglected if the goal is to increase effectiveness of FITs.
Keywords: Photovoltaic solar, Feed‐in tariffs, Policy‐consistency, Design characteristics
JEL‐codes: C23, G11, H23, N70, Q42, Q48
*
Corresponding author. E‐mail address: dijkgraaf@ese.eur.nl (E. Dijkgraaf).
1
1. Introduction
In the last decades an increasing number of governments started to stimulate the
development of renewable electricity sources. Important motivations have been the growing
concern for climate change and the rising scarcity of fossil fuels. Also international agreements
on carbon emission reduction and renewable energy targets and governments’ objectives to
reduce their dependence on energy imported from abroad played a role.
From the most common sources of renewable electricity – biomass, wind, solar, geothermal
and hydropower – photovoltaic solar (PV) has made its growth spurt most recently, with a
worldwide average yearly growth of 47% over the last decade. This development has
increasingly resulted in up‐scaled manufacturing facilities and technology improvements. As
both have driven the price of a PV‐system down, PV has gradually become a competitive
electricity source. At the start of the development of the PV‐industry in the 1990s, important
conventional electricity sources as nuclear and coal‐fired power plants were able to produce
electricity at only a fraction of the costs at which one solar kilowatt‐hour (kWh) could be
generated. For bridging the competitive gap between conventional and renewable electricity
sources, the role of governments has generally been considered as crucial.
The development of PV started in the early 90s in Germany and Japan; these two countries
were the first to adopt policy‐instruments. Since 2005, governments all around the world have
started to promote PV. Policy‐makers have introduced several combinations of instruments.
The most common instrument is a Feed‐in Tariff (FIT). The general idea of a FIT is that the
owner of a PV‐system receives on contract basis a guaranteed price for every produced kWh
that is fed into the grid during a fixed period of time. FITs can differ in various characteristics,
including tariff amount, limitations on available budget or installed capacity and contract
duration.
In the past few years uncertainty increased about how effective FIT policies have been in the
development of PV. Policy‐makers also have difficulty finding the optimal structure of a FIT.
Lacking the right FIT‐structure can be the reason that some countries are lagging behind in the
development compared to others. In this light, it is worthwhile to analyze the relationship
between effectiveness and structure of a FIT. Besides FIT‐structure, policy‐consistency also
2
influences the effectiveness of a FIT (see White et al., 2013, Owens & Driffill, 2008, Gardner &
Stern, 1996). Investors seek maximum returns at an acceptable risk. If government support is
vital to provide a reasonable return, policy‐consistency is considered as crucial to limit the risk.
Investors often complain about the regular changes in policies as they have to decide about
multi‐year investment plans. If the lifespan of policies and investments differ significantly, risk
increases for investors.
This paper studies whether FIT policies have been effective in the development of solar PV. To
the best of our knowledge, this is the first paper empirically taking into account design
characteristics and consistency of FIT policies. The majority of the existing literature is limited
to a descriptive approach in analyzing the effectiveness of policy instruments on the
development of renewable energy sources. Haas et al. (2011), who compare quantity‐driven
(like tradable green certificates based on quotas) and price‐driven instruments (like FITs) in
seven European countries, state that FIT policy instruments are more effective than others,
because they usually have low administration costs, are relatively easy to implement and are
technology‐focused. They underpin that a well‐designed FIT policy provides deployment of
renewable energy sources in the shortest time and at the lowest costs for society. Also Gipe
(2006), Mendonça (2007), Mendonça et al. (2009), Cory et al. (2009) and Timilsina et al. (2012)
show qualitatively that FITs are a major driver for the development of most solar PV markets.
Jenner et al. (2013) and Bürer and Wustenhagen (2009) add that investors prefer a FIT over
other policy instruments. These qualitative policy evaluation studies generally agree that FITs
are important in explaining the development of renewable energy sources in Europe. Stronger
FITs with higher returns on investment usually have a bigger impact on the development.
Jenner et al. (2013) and Jenner (2012) are the only two empirical studies that have assessed
the effectiveness of FITs in promoting the development of PV. Jenner et al. (2013) develop an
indicator for FIT strength that captures several design characteristics such as tariff size,
contract duration, degression rate but also electricity price and production cost to estimate
the resulting return on investment (ROI). They find that for a 10% increase in ROI, 3.8% more
PV capacity will be installed on average per year. All together their conclusion is that FIT
policies have stimulated the development of PV in Europe between 1992 and 2008. Jenner
(2012), who uses generation in GWh instead of annual solar capacity as dependent variable,
3
comes to the same conclusion for the period 1990‐2010. An important shortcoming of these
studies is that both the individual design characteristics of a FIT and the price of electricity are
not included as separate variables in the regression. Therefore it is hard to determine the
impact of the individual elements. Besides that the role of policy‐consistency is ignored, just
as the possible presence of a cap on cost or capacity. Our study does include these indicators.
Moreover, in contrast to the above two studies, our study also captures tariff differences due
to installation size (which is a crucial determinant to measure the effectiveness of a FIT
correctly).
A related paper, Marques & Fuinhas (2012), analyzes the impact of policy incentives on the
total contribution of renewables to total energy supply (and not on the development of solar
PV separately) by focusing on 24 European countries. Their results give empirical support for
the assumption that public policy measures contribute to the wider use of renewables. In
particular direct interventions like incentive and subsidy policies, including FITs, turned out to
be effective. Quota obligations, R&D programs and tradable certificates did not increase
renewables in the period under study. They also find that strategic planning processes
contribute to the PV development. This study does have shortcomings though. All policy
variables are composed by counting the number of active policies of that policy type in a
specific year and country. This method does not capture any of the heterogeneity in the policy
design and does not differentiate between dissimilar market circumstances. Also, only data
until 2007 are used, while the development of PV is dominated by more recent years.
The rest of the paper is organized as follows. The next section specifies the structure of the
model and describes the data set and variables used. Section 3 gives an overview of the
different results and provides an interpretation. Finally, Section 4 concludes.
4
2. Models and data
2.1 The models
In order to investigate the impact of FITs on the development of PV a series of linear models
are investigated.
The specification of the models is as follows:
PVit = β1 + β2jt FITit + β3ktZit + β4tTt + ηi + εit, (1)
where PVit is a measure for the use of PV in country i in year t, FIT is a vector of (combinations)
of the strength of a FIT (FITSTRENGTH), policy consistency measures and underlying
characteristics of the FIT, Z is a vector of other characteristics, T is a time trend, ηi are country
fixed effects and εit is the error term.
PV is measured both as a share in the production of electricity and as added yearly capacity
per capita. The first is interesting to throw light on the development of the PV‐share, while
the second is far more directly related to yearly developments.
FITSTRENGTH is approximated by inventorying the underlying characteristics of the FIT as
simple counting whether a FIT is present cannot answer our research questions. This makes it
not only possible to have a more precise measure for a FIT, but also to explicitly include the
underlying characteristics of the FITs to see whether FIT design matters. We add measures for
policy consistency by measuring FITSTRENGTH as the average for several years (if the FIT is
high for more years, the average will be higher) and by adding the standard deviation of
FITSTRENGTH (if the changes in the FIT are large, the standard deviation will be higher). We
expect positive signs, except for the standard deviation (a higher deviation points to less policy
consistency).
We correct for several characteristics that differ between countries and years. We discuss this
in the data section.
5
We have included a time‐trend to correct for autonomic developments in our analysis, like
technological developments. Part of the development of the PV‐industry might be caused by
these autonomic developments and not by FITSTRENGTH or other variables that are included
in the model; the time‐trend adjusts for this potential bias.
Country specific effects are included to correct for, for example, differences in availability of
sun hours.
2.2 Data collection
Yearly data are collected for 30 OECD member countries over the period 1990‐2011, which
sums up to 660 observations. Four OECD member countries are not included in our panel:
Iceland, Canada, Australia and the USA. Iceland is not taken into account because of the fact
that data was lacking on electricity prices. Canada, Australia and the USA are all excluded for
the reason that their electricity markets are divided in regions or states with the consequence
that different policies are implemented within one country. Since we perform country‐level
analysis it is not possible to include these countries. The data are primarily obtained from the
World Bank, OECD‐iLibrary and IEA databases. Other sources are mentioned in the specific
data sections.
Some variables had missing observations. Since we had data for two electricity price variables
(for households and the industrial segment), we used the growth rate of one variable if
observations for the other variable were missing. For country‐years where both variables had
missing observations, we used the average growth‐rate of five consecutive years in the direct
past or future. When no policy data was available for a specific country and the development
of the share in the production of electricity (PVSHARE) was zero or very close to zero, we
assumed that no policy had been in place in this specific country.
2.3 Measuring PV
As already mentioned above PV is measured both as a share in the production of electricity
(PVSHARE) and as added yearly capacity per capita (ADDCAPC). Both provide interesting
information. PV as a share of electricity production shows whether the goal of many countries,
6
i.e. increasing the share of renewable energy, is reached and what the contribution of FITs is
to this goal. Yearly added capacity is also interesting as it measures more directly what the
development of PV is.
Having a ratio (PVSHARE) as dependent variable is in contrast to most existing empirical
literature where absolute values are chosen. Jenner et al. (2013), for example, have chosen
for capacity to reflect the investment decision as purely as possible. A disadvantage of
absolute values (e.g. cumulative or added PV capacity) is that the results are dominated by
bigger countries which results in heteroskedasticity. We measure added capacity therefore
per capita. In a sensitivity analysis we test whether replacing PVSHARE by the quantity of
produced electricity per capita changes our conclusions.
Table 1
Descriptive statistics dependent variables.
Mean Std.Dev. Minimum Maximum
PVSHARE 0.06 0.30 0.00 3.57
ADDCAPC 2.05 10.73 0.00 153.09
Table 1 shows the descriptive statistics of the dependent variables. PVSHARE is still limited
with a mean of only 0.06% over the whole panel. Important to note is that until 2004 only one
observation has a PVSHARE above 0.06%, i.e. Japan in 2003. In contrast, the average PVSHARE
for all observations in the last five years of our dataset is 0.24% and in 2011 even 0.61%. Figure
1 gives an overview of the development of the average PVSHARE in our dataset. Additionally
it shows the number of countries with a FIT in place. The figure clearly shows that the share
of PV strongly grew in the last few years. Part of this growth might be explained by the
increasing number of active FITs. PVSHARE increased significantly in a few forefront countries
as Germany (2008: 0.7%, 2011: 3.1%), Spain (2008: 0.8%, 2011: 2.8%) and Italy (2008: 0.06%,
2011: 3.6%).
7
Figure 1. Average PVSHARE and number of countries with FIT
Figure 2. Average ADDCAPC and number of countries with FIT
2.4 Measuring FIT
FITs are evidently the most popular mechanism to promote the PV‐industry within our
country‐panel. Out of 30 OECD member countries included in this study 22 have implemented
a FIT. In 20% of the observations a FIT was in place; in the last decade of our data‐period (2002‐
2011) even 43% of the years include a FIT‐policy.
8
To measure the strength of a FIT we distinguish four strength measures (see Table 2): tariff
amount (split into three parts; tariffs for small, commercial and utility scale systems), contract
duration, limitation on cost or capacity and average cost of generating one solar kWh for each
panel observation (taking solar irradiation and PV‐system price into account).
To be able to combine the four measures we reward each of them with points on a pre‐defined
scale. The sum of the rewarded points forms the FIT‐strength indicator. This is a transparent
alternative for the ROI indicator that is composed by Jenner et al. (2013). By rewarding the
four FIT design characteristics individually we are able to test the effectiveness of these
characteristics separately as well as combined as one FIT‐strength indicator.
Table 2
Points classification of FIT strength measures.
Nr. FIT strength measure Points
1a FIT‐Tariff amount: small systems [0,25]
1b FIT‐Tariff amount: commercial systems [0,25]
1c FIT‐Tariff amount: utility scale systems [0,25]
2 FIT‐Duration contract [‐20,20]
3 FIT‐Cap on cost or capacity Yes = ‐15, No = 0
4 FIT‐Cost of generation [‐20,20]
Table 3
Descriptive statistics for FIT measures.
Mean Std.Dev. Minimum Maximum
FIT‐Tariff Small 11.87 5.26 1.00 20.00
FIT‐Tariff Commercial 10.98 5.31 0.00 20.00
FIT‐Tariff Utility 7.00 7.39 0.00 22.00
FIT‐Tariff 9.95 4.98 0.67 20.33
FIT‐Duration 4.03 8.84 ‐10.00 20.00
FIT‐Cap ‐4.03 6.68 ‐15.00 0.00
FIT‐Cost 13.12 3.57 5.00 19.00
FITSTRENGTH 42.94 20.65 0.00 82.00
Note: descriptive statistics are given for subsample of observations with a FIT (130).
9
Rewarding each of the four measures properly is crucial; both knowledge of actual data and a
reasonable division of points are required to obtain a well‐balanced and correct FIT‐strength
indicator. Table 2 gives an overview of the scales on which points are assigned to each
measure.1 The chosen points ratio between one measure and the other is based on personal
calculations and in‐depth market knowledge. Table 3 displays the descriptive statistics of the
involved measures. Data on the FIT policies have been obtained from a variety of sources
including PV Grid, Wind‐works.org, IEA Policies & Measures Databases, RES Legal and Europe’s
Energy Portal. A detailed database on FIT policies does not exist but is composed by the
authors. We have categorized the data such that they could reasonably be compared between
countries and years. The following illustrates our approach.
Since all measures have only values in the years that a FIT was in place, in Table 3 the
descriptive statistics for only these 130 observations are shown. The minimum‐maximum
ranges of points that are actually rewarded, shown in Table 3, deviate somewhat from the
classification in Table 2 for some of the measures. This is mainly the case for the measures
contract duration (FIT‐DURATION) and Average Cost of Generation (ACG) of one kWh (FIT‐
COST). The reason for these deviations is that countries in our data panel did not implement
FITs with a contract of less than 10 years2 and that no country implemented a FIT when the
average cost of generation was above 25.4 eurocent.3
The division of points is linear and based on the actual minimum and maximum values of each
measure. Let’s illustrate this with respect to the tariff amount measure. The lowest tariff
amount for small PV‐systems (Measure 1a in Table 2) is 5.2 eurocent, the highest is 60
eurocent. Points are divided linearly between 0 eurocent (0 points) and 75 eurocent (25
points) (see Table A1 in Appendix A). A tariff amount between 3 and 6 eurocent is rewarded
1 point, a tariff amount between 60 and 63 eurocent is rewarded 20 points. Thus points are
assigned in the range 1‐20. The same method is applied for tariff amounts of commercial and
utility scale systems. For FIT‐DURATION and FIT‐COST we have awarded points in the range of
1
Table A1 in Appendix A shows a more detailed overview of how the points are assigned.
2
A contract of 10 years corresponds with ‐10 points in Table A1 in Appendix A.
3
An average cost of generation of 25.4 eurocent corresponds with 5 points in Table A1 in Appendix A.
10
‐20 points (5 year) to 20 points (25 year) respectively 5 points (25.4 eurocent) to 19 points (3.8
eurocent).
As we have mentioned above, the tariff amount measure is split into three different
categories. Since we expect the tariff amount to be more effective for larger systems (due to
the fact that the life cost of energy of larger systems is lower), we have rewarded the tariff
amount of larger systems with more points. We assumed that small systems need to have on
average 10% higher tariff than commercial systems and 20% higher than utility scale systems
to be equally effective.
FIT‐DURATION has been rewarded with both negative and positive points. A FIT that includes
a contract of more than 15 year is generally considered as attractive since tariffs are paid over
a period more comparable with the lifetime of a PV‐system. A balanced trade‐off between
contract duration and tariff amount is important to retain a profitable ROI. As not many
investors have been able to obtain a good ROI with a short contract, a contract of less than 15
years has been rewarded with negative points.
A limitation on cost or capacity, also known as a cap (FIT‐CAP), is a pre‐defined maximum on
either the budget or the installed capacity during a fixed period of time. A cap can cause
uncertainty to potential investors as they are no longer sure whether pre‐investment costs
can be earned back. As a cap has a negative impact on the effectiveness of a FIT, negative
points are awarded if a cap is included. We use a binary valuation for this specific design
characteristic since – due to the high variability in caps – it is hard to classify the caps and since
there are relatively few FIT policies that include a cap. The inclusion of a cap (no matter which
one) is rewarded with ‐15 points.
As shown in Table 2, a fourth measure is included: the average cost of generation (FIT‐COST).
Two countries that implement FITs with equal tariff amount and contract duration will not
necessarily have a comparable development; ROI is based on more factors than these two
alone. ROI varies primarily because of the difference in insolation4 between countries. Purely
4
The intensity of incoming solar radiation incident on a square meter horizontal surface.
11
based on solar irradiation, generation in Israel is about 70% higher than the generation of a
PV‐system in the average country in our panel. But price, lifetime and efficiency of a PV‐system
also contribute to its variation. By keeping all other factors constant (lifetime = 25 years, PV‐
system efficiency = 15% and one sq. meter of solar panels consist of 150 Watt peak (Wp)), we
have calculated the average number of kWh’s that is generated by a system per Wp for each
country‐year. Dividing the prevailing PV‐system price per Wp by that number, gives us the
average cost of generating one kWh of solar electricity over the lifetime of a PV‐system. The
lower this number, the higher the reward in points is.
The unweighted sum of the rewards forms the strength of the FIT (FITSTRENGTH) in our main
analysis. In sensitivity analyses we give subsequently more weight to each underlying
characteristic to test whether our results are robust.
The dots in Figure 3 are showing all FITSTRENGTH observations through the years. FITs become
on average stronger through the years. Part of this increase is of course explained by
decreasing system prices which makes PV more and more attractive. As an example, the
development of FITSTRENGTH is highlighted for Germany (solid line), Spain (line with small
dots) and Italy (line with bars). The development is very consistent over the years in Germany,
but far less so in Italy (big jump in 2008) and Spain (jump in 2008, but sharp decline in 2009
and 2011). This indicates that variation in the panel is available, not only with respect to
FITSTRENGTH levels between countries, but also to changes over time.
We are not only interested in the effects of FIT and its design characteristics, but also in the
effects of policy consistency. The hypothesis is that the effects of a FIT are larger when a
country has a more consistent policy over time. If potential investors gain confidence in the
continuity of a policy, the effectiveness of a FIT is expected to increase. So the higher and more
constant tariffs over several years are, the higher the level of consistency, the higher the
expected effectiveness.
We measure consistency by two variables. First, we replace FITSTRENGTH by the average FIT
over the past 5 years (FITSTRENGTH(5)). As a sensitivity analysis we do also estimations with
12
the average over the last 2, 4, 6, 8 and 10 years. Second, we add to the FITSTRENGTH variable
the standard deviation of FITSTRENGTH (FITSTRENGTH(SD)) over the past 5 years.
Descriptive statistics of the FIT variables are shown in Table 4.
Figure 3. Overview FITSTRENGTH observations and development three countries.
Table 4
Descriptive statistics FIT‐policy variables.
Mean Std.Dev. Minimum Maximum
FITSTRENGTH 8.46 19.38 0.00 82.00
FITSTRENGTH(5) 4.39 12.92 0.00 72.00
FITSTRENGTH(SD) 2.95 7.45 0.00 42.46
2.5 Other policy variables
As a sensitivity analysis we also include the following other policy variables:
RD&D budgets for (governments) investments in the PV sector (RDENDBUDGET);
investment and tax incentives (POLINVEST);
net metering policies (POLNETMET);
calls for tender (POLTEN).
13
We measure RD&D budgets in million US$ and the other three policies simply through binary
variables, by counting whether a certain policy is present or not. We include these policies
only in a sensitivity analysis since there are not many observations per policy available and
since three of these policies have an imprecise binary measure. We leave a more robust
analysis of these policies for further research.
Descriptive statistics of the other policy variables are shown in Table 5.
Table 5
Descriptive statistics other policy variables.
Mean Std.Dev. Minimum Maximum
RDENDBUGDET 9.49 24.21 0.00 225.74
POLINVEST 0.29 0.45 0.00 1.00
POLNETMET 0.06 0.24 0.00 1.00
POLTEN 0.01 0.09 0.00 1.00
2.6 Exogenous and control variables
The following exogenous variables are included to control for differences between countries
and in time:
wealth (both GDPcap and GDPcap2) to account for a possible U‐shaped curve
analogous to the Environmental Kuznets Curve (see Grossman and Krueger, 1991,
Panayotou, 1993);5
trade openness of a country (OPENNESS) measured as the percentage of export and
import of GDP as more open countries might easier import new PV technologies and
profit more from PV export, making national investments more profitable;
population density (POPDENS) measured as the number of inhabitants per squared
kilometer as PV is a land intensive energy technology;
energy import (ENERGYIMPORT) measured as share of total energy as a higher share
might give more incentives to decrease this share by investing in PV;
5
The Environmental Kuznets Curve hypothesis postulates an inverted‐U‐shaped relationship between income
levels and negative environmental effects, which might implicate a U‐shaped relationship between income levels
and PV.
14
the capacity of the electricity plants (ELECCAPACITY) as a higher capacity results in
more replacement investments and thus a higher chance of investments in PV;
the price of electricity (ELECPRICEHH) as a higher price makes it more likely that the
business case for PV becomes profitable. We do not include industrial electricity prices
because this would cause collinearity and would not contribute to the explanatory
power of the model. We have chosen for the household electricity price variable
because it is less correlated with the other variables in the model.
Data for GDPcap, OPENNESS, and POPDENS are obtained from the World Bank.
ENERGYIMPORT, ELECCAPACITY, and ELECPRICEHH data are all derived from the OECD‐
iLibrary. Descriptive statistics of both exogenous and control variables are given in Table 6.
Table 6
Descriptive statistics control variables.
Mean Std.Dev. Minimum Maximum
GDPcap 18.20 11.46 1.65 56.39
OPENNESS 87.66 48.44 15.92 319.55
POPDENS 149.05 125.74 13.09 511.37
ENERGYIMPORT 25.90 135.74 ‐842.00 99.00
ELECCAPACITY 1.65 1.22 0.00 7.17
ELECPRICEHH 137.46 64.38 10.32 409.17
15
3. Results
This section discusses the empirical results.6 First we discuss the effectiveness of the general
FIT measure and the impact of the consistency of FITs. Next, we delve into the relation
between effectiveness and design characteristics of the FIT. Finally, we present some
sensitivity analyses.
3.1. Effectiveness of FIT
Table 7 presents the estimation results for the base models. In all cases FITSTRENGTH has a
positive and significant (at 1%) effect on the production of PV, measured as share of produced
electricity. If a country introduces a FIT with a strength of 82 (the maximum in the sample) the
effect is +0.2% on the PV share in electricity production. Compared with the mean PV share
of 0.06% this is a relative large effect. The effect becomes 5.5 times higher if the indicator is
changed to the average of FITSTRENGTH in the preceding 5 years. Estimations with other
averages (2, 4, 6, 8, 10 years) show that the effectiveness is linearly increasing in the lag period
(see Table 8 and 9). For 10 years the effectiveness is more than 10 times as high with an effect
of 2.14% on the production of PV. This means that effectiveness increases considerably if over
a longer period of time FITSTRENGTH is high. If the standard deviation is included in the
estimations, the effect of FITSTRENGTH is also much higher, but decreases if the standard
deviation increases. Again, this indicates that consistency is very important for the effect of
FIT on the production of PV.
These effects are reproduced if added capacity per capita is taken as endogenous variable.
The effect of FITSTRENGTH is positive and significant and increasing if it is higher over a longer
period and decreasing if the standard deviation is larger. The effect of consistency measured
as average over the previous 5 years for produced solar electricity is twice the effect for added
capacity. This is not only the case for 5 years, but also for 2, 4, 6, 8 and 10 years. On the other
hand, if the standard deviation is used as measure for consistency the effect of this variable is
larger for added capacity than for produced solar electricity.
6
As the Breusch Pagan test is rejected for all variables on 1%, standard errors are White corrected.
16
The result for GDP per capita results in a non‐linear relation with the production electricity
share of PV. The effect of GDP growth is negative until a level of 37.000 dollars per inhabitant
and positive above that level. For added capacity the results are quantitative comparable, but
they are less significant. Openness has no influence on the production or added capacity of
PV. Population density has a negative influence in most estimations. This is not surprising as
PV is a land intensive way of producing electricity. The influence of energy imports is not
significant. For the electricity capacity and the price of electricity we find clearly a positive
effect. With a high capacity more new investments are needed, giving space to the choice of
PV. A higher price of electricity clearly stimulates relative expensive solutions like PV as the
business case of PV will be more often positive.
Table 7
Estimation results base model.
PVSHARE ADDCAPC
Current 5 yr St dev Current 5 yr St dev
FITSTRENGTH 0.002*** ‐ 0.011*** 0.183*** ‐ 0.557***
(0.20) (0.87) (14.98) (45.65)
FITSTRENGTH(5) ‐ 0.013*** ‐ ‐ 0.483*** ‐
(1.10) (39.57)
FITSTRENGTH(SD) ‐ ‐ ‐0.022*** ‐ ‐ ‐1.027***
(‐0.33) (‐15.27)
GDPcap ‐0.064*** ‐0.052*** ‐0.056*** ‐1.475* ‐1.172* ‐1.068
GDPcap2 0.001*** 0.001*** 0.001*** 0.017** 0.016** 0.015
OPENNESS 0.001 0.000 0.001 0.020 0.003 0.013
POPDENS ‐0.001*** ‐0.001* 0.000 ‐0.051** ‐0.031* ‐0.007
ENERGYIMPORT ‐0.001 0.000 0.000 ‐0.016 ‐0.004 ‐0.006
ELECCAPACITY 0.016*** 0.010*** 0.014*** 0.348** 0.190* 0.266**
ELECPRICEHH 0.002** 0.001* 0.002* 0.057*** 0.029* 0.039*
TREND ‐0.006 ‐0.008 ‐0.006 ‐0.166 ‐0.125 ‐0.191*
Constant 0.198 0.316* 0.047 7.670 8.951 0.746
Fixed effects Yes Yes Yes Yes Yes Yes
Observations 660 660 660 660 660 660
R2 0.39 0.57 0.50 0.30 0.42 0.46
Notes: */**/*** means significance at 10%/5%/1%. Marginal effects based on maximum FITSTRENGTH (82)
and average standard deviation FITSTRENGTH (14.86) between brackets below coefficients. To correct for
heteroscedasticity standard errors are White corrected.
17
Table 8
Estimation results 2, 4, 6, 8 and 10 years PVSHARE.
2 yr 4 yr 6 yr 8 yr 10 yr
FITSTRENGTH(5) 0.007*** 0.011*** 0.016*** 0.021*** 0.026***
(0.58) (0.91) (1.32) (1.69) (2.14)
GDPcap ‐0.057*** ‐0.053*** ‐0.051*** ‐0.052*** ‐0.052***
GDPcap2 0.001*** 0.001*** 0.001*** 0.001*** 0.001***
OPENNESS 0.001 0.001 0.000 0.000 0.000
POPDENS ‐0.002* ‐0.002 ‐0.001 ‐0.001 ‐0.001
ENERGYIMPORT 0.000 0.000 0.000 0.000 0.000
ELECCAPACITY 0.013*** 0.011*** 0.009*** 0.009*** 0.008***
ELECPRICEHH 0.002*** 0.001*** 0.001*** 0.001*** 0.001**
TREND ‐0.008* ‐0.008** ‐0.007* ‐0.006* ‐0.005
Constant 0.293 0.303 0.324 0.330* 0.354*
Fixed effects Yes Yes Yes Yes Yes
Observations 660 660 660 660 660
R2 0.46 0.52 0.58 0.59 0.61
Table 9
Estimation results 2, 4, 6, 8 and 10 years ADDCAPC.
2 yr 4 yr 6 yr 8 yr 10 yr
FITSTRENGTH(5) 0.304*** 0.419*** 0.571*** 0.763*** 0.958***
(24.95) (34.33) (46.81) (62.56) (78.55)
GDPcap ‐1.285** ‐1.195** ‐1.143** ‐1.133** ‐1.164**
GDPcap2 0.017** 0.017** 0.016** 0.016** 0.016***
OPENNESS 0.011 0.007 0.000 ‐0.005 ‐0.008
POPDENS ‐0.052 ‐0.038 ‐0.025 ‐0.014 ‐0.008
ENERGYIMPORT ‐0.009 ‐0.005 ‐0.003 ‐0.001 0.000
ELECCAPACITY 0.271*** 0.214*** 0.163*** 0.125*** 0.101**
ELECPRICEHH 0.043*** 0.034*** 0.023** 0.015 0.011
TREND ‐0.164 ‐0.149 ‐0.102 ‐0.064 ‐0.028
Constant 9.311 8.818 9.165 9.698 10.504
Fixed effects Yes Yes Yes Yes Yes
Observations 660 660 660 660 660
R2 0.36 0.40 0.44 0.47 0.49
3.2. Effectiveness of separate FIT design characteristics
Table 10 presents the estimation results for the FIT design characteristics. Nearly all
characteristics are positive and significant. Only for the effect of capacity restriction on the
production share of PV no effect is found. The largest effect is found for the tariff. With the
highest available tariff in our sample, the marginal effect of FIT is 0.5% on the production
share. For this share the second most effective characteristic is cost, followed by duration. For
added capacity tariff is also the most effective characteristic. The other three characteristics
18
have comparable effects. We can conclude that tariff is the most important design
characteristic, but that the effectiveness is largest when all characteristics are optimized.
Table 10
Estimation results FIT design characteristics.
PVSHARE ADDCAPC
FITSTRENGTH(SD) ‐0.018*** ‐0.820***
(‐0.27) (‐12.19)
FIT‐TARIFF 0.024*** 1.785***
(0.50) (36.30)
FIT‐COST 0.018*** 0.629***
(0.37) (12.57)
FIT‐DURATION 0.013* 0.788***
(0.25) (14.98)
FIT‐CAP 0.001 0.753***
(0.01) (11.29)
GDPcap ‐0.059*** ‐1.390
GDPcap2 0.001*** 0.014
OPENNESS 0.001 0.020
POPDENS 0.000 0.071*
ENERGYIMPORT 0.000 ‐0.007
ELECCAPACITY 0.014** 0.320*
ELECPRICEHH 0.002** 0.056**
TREND ‐0.015** ‐0.824***
Constant 0.069 ‐7.785*
Fixed effects Yes Yes
Observations 660 660
R2 0.49 0.45
Notes: */**/*** means significance at 10%/5%/1%. Marginal effects based on
average standard deviation FITSTRENGTH (14.86), maximum for TARIFF (20.33),
COST (20), DURATION (19), and minimum (in absolute value) for CAP between
brackets below coefficients. To correct for heteroscedasticity standard errors are
White corrected.
3.3. Sensitivity analyses
In Section 3.1 and 3.2 we have found that the FIT is effective in stimulating the production and
capacity of PV and that all design characteristics are important if the goal is to increase the
effectiveness of the FIT. In this section we show whether these conclusions are affected when
we change several assumptions.
First, we changed the weighting schemes of the underlying parts of FITSTRENGTH. The
underlying characteristics of the FITs are weighted differently compared with the base model
(the model with FITSTRENGTH and its standard deviation). In each sensitivity analysis the
19
weight of respectively tariff, cost, duration and cap is doubled. In all cases FITSTRENGTH and
its standard deviation are significant and the sign is the same as in the base model (see Table
11 and 12). Marginal effects are comparable.
Table 11
Estimation results PVSHARE: different weighting schemes FIT.
Double weight for:
Base Tariffs Cost Duration Cap
FITSTRENGTH 0.011*** 0.005*** 0.009*** 0.010*** 0.007***
(0.87) (0.67) (0.85) (0.98) (0.57)
FITSTRENGTH(SD) ‐0.022*** ‐0.011*** ‐0.018*** ‐0.020*** ‐0.013***
(‐0.33) (‐0.26) (‐0.34) (‐0.33) (‐0.19)
Fixed effects Yes Yes Yes Yes Yes
Observations 660 660 660 660 660
R2 0.50 0.44 0.49 0.48 0.42
Notes: */**/*** means significance at 10%/5%/1%. Estimation results for other variables (models are
comparable with Table 7 and 10) available on request. Marginal effects based on maximum FITSTRENGTH
and average standard deviation FITSTRENGTH between brackets below coefficients. To correct for
heteroscedasticity standard errors are White corrected.
Table 12
Estimation results ADDCAPC: different weighting schemes FIT.
Double weight for:
Base Tariff Cost Duration Cap
FITSTRENGTH 0.557*** 0.283*** 0.400*** 0.430*** 0.525***
(45.65) (38.43) (38.83) (43.86) (43.08)
FITSTRENGTH(SD) ‐1.027*** ‐0.522*** ‐0.721*** ‐0.759*** ‐0.947***
(‐15.27) (‐13.01) (‐13.47) (‐12.27) (‐13.53)
Fixed effects Yes Yes Yes Yes Yes
Observations 660 660 660 660 660
R2 0.50 0.44 0.49 0.48 0.42
Notes: */**/*** means significance at 10%/5%/1%. Estimation results for other variables (models are
comparable with Table 7 and 10) available on request. Marginal effects based on maximum FITSTRENGTH
and average standard deviation FITSTRENGTH between brackets below coefficients. To correct for
heteroscedasticity standard errors are White corrected.
Second, we replaced PV as share of electricity production by produced electricity per capita.
Again we find that the effect of FIT is significant, positive and increasing in consistency (see
Table 13). The results for the design characteristics are also comparable, although cost and
duration do have different significance levels now.
20
Table 13
Estimation results FIT production electricity per capita.
Current 5 years St.dev. Design
FITSTRENGTH 1.892** ‐ 7.596*** ‐
(155.12) (622.83)
FITSTRENGTH(5) ‐ 9.353*** ‐ ‐
(766.99)
FITSTRENGTH(SD) ‐ ‐ ‐15.666*** ‐12.342***
(‐232.80) (‐183.40)
FIT‐TARIFF 18.064***
(367.24)
FIT‐ COST 12.252*
(245.04)
FIT‐DURATION 10.634***
(202.04)
FIT‐CAP 1.084
(16.25)
Fixed effects Yes Yes Yes Yes
Observations 660 660 660 660
R2 0.40 0.57 0.49 0.47
Notes: */**/*** means significance at 10%/5%/1%. Estimation results for other variables (models are
comparable with Table 7 and 10) available on request. Marginal effects based on maximum FITSTRENGTH (82)
and average standard deviation FITSTRENGTH (14.86) between brackets below coefficients. To correct for
heteroscedasticity standard errors are White corrected.
Third, we estimated the effect of FIT including variables for other policy instruments (RD&D
budgets for investments in the PV sector, investment and tax incentives, net metering policies,
calls for tender). As these instruments are less often used, resulting in few observations, and
are measured in a rough manner (based only on a simple count of available instruments), we
do not include these results in our base models. Most important is that including the four
other policy instruments does not influence our main conclusions (see Table 14). The effects
of the other instruments are in all cases insignificant, while the coefficients for FITSTRENGTH
and its standard deviation are highly comparable.
21
Table 14
Estimation results other instruments
PVSHARE ADDCAPC
FITSTRENGTH 0.011*** 0.555***
(0.89) (45.51)
FITSTRENGTH(SD) ‐0.023*** ‐1.029***
(‐0.34) (‐15.29)
POLINVEST ‐0.022 0.729
POLNETMET ‐0.026 1.821
POLTEN ‐0.063 ‐0.879
RDENDBUDGET ‐0.001 ‐0.027
GDPcap ‐0.054** ‐1.128
GDPcap2 0.001*** 0.015*
OPENNESS 0.001 0.010
POPDENS 0.000 ‐0.006
ENERGYIMPORT 0.000 ‐0.005
ELECCAPACITY 0.014*** 0.271**
ELECPRICEHH 0.002* 0.039*
TREND ‐0.006 ‐0.198*
Constant 0.000 1.600
Fixed effects Yes Yes
Observations 660 660
R2 0.47 0.42
Notes: */**/*** means significance at 10%/5%/1%..Marginal effects
based on maximum FITSTRENGTH (82) and average standard
deviation FITSTRENGTH (14.86) between brackets below coefficients.
To correct for heteroscedasticity standard errors are White corrected.
22
4. Conclusions
This paper has empirically analyzed whether FIT policies have been effective in the
development of solar PV. In contrast to the existing literature FIT‐structure and policy‐
consistency were explicitly taken into account.
We have found a strong positive relation between the presence of a FIT and the development
of a country’s share of PV in the electricity‐mix. This effect is increasing if the FIT policy is more
consistent. The tariff of the FIT is the most important design characteristic; the other
characteristics (cost levels, duration of contracts and caps w.r.t. capacity or costs) are
important too though. If the goal is to maximize the effectiveness of FITs the design
characteristics should be consistent in time and optimized for each relevant aspect.
We have indications that our results are robust. Measuring FIT policies in other ways, (by
changing the weighting schemes of the underlying characteristics), changing the endogenous
variable, varying the time scale of measuring, and including other policy instruments did not
change our results. Especially with respect to the sensitivity analysis with non‐FIT policy
instruments included, we feel that it might be interesting to invest in new research as our non‐
FIT policy variables are based on rough measures only.
23
Appendix A
Table A1
Detailed overview of points per FIT measure.
FIT‐TARIFF FIT‐DURATION FIT‐COST
Points Small (€) Comm (€) Utility (€) Years Points ACG (€) Points
0 0 0 0 5 ‐20 0.62 ‐20
1 0.03 0.027 0.024 6 ‐18 0.605 ‐19
2 0.06 0.054 0.048 7 ‐16 0.59 ‐18
3 0.09 0.081 0.072 8 ‐14 0.575 ‐17
4 0.12 0.108 0.096 9 ‐12 0.56 ‐16
5 0.15 0.135 0.12 10 ‐10 0.545 ‐15
6 0.18 0.162 0.144 11 ‐8 0.53 ‐14
7 0.21 0.189 0.168 12 ‐6 0.515 ‐13
8 0.24 0.216 0.192 13 ‐4 0.5 ‐12
9 0.27 0.243 0.216 14 ‐2 0.485 ‐11
10 0.3 0.27 0.24 15 0 0.47 ‐10
11 0.33 0.297 0.264 16 2 0.455 ‐9
12 0.36 0.324 0.288 17 4 0.44 ‐8
13 0.39 0.351 0.312 18 6 0.425 ‐7
14 0.42 0.378 0.336 19 8 0.41 ‐6
15 0.45 0.405 0.36 20 10 0.395 ‐5
16 0.48 0.432 0.384 21 12 0.38 ‐4
17 0.51 0.459 0.408 22 14 0.365 ‐3
18 0.54 0.486 0.432 23 16 0.35 ‐2
19 0.57 0.513 0.456 24 18 0.335 ‐1
20 0.6 0.54 0.48 25 20 0.32 0
21 0.63 0.567 0.504 0.305 1
22 0.66 0.594 0.528 FIT‐CAP 0.29 2
23 0.69 0.621 0.552 Cap? Points 0.275 3
24 0.72 0.648 0.576 Yes ‐15 0.26 4
25 0.75 0.675 0.6 No 0 0.245 5
0.23 6
0.215 7
0.2 8
0.185 9
0.17 10
0.155 11
0.14 12
0.125 13
0.11 14
0.095 15
0.08 16
0.065 17
0.05 18
0.035 19
0.02 20
24
Table A2. List of variables
PVSHARE Share of electricity generated with PV in %
total generated electricity
ADDCAPC This year installed new capacity for PV per MWp
capita
FITSTRENGTH Measure for strength feed‐in tariff this year Points
FITSTRENGTH(5) Measure for strength feed‐in tariff average Points
of last five years
FITSTRENGTH(SD) Measure for strength feed‐in tariff: Points
standard deviation of last five years of
FITSTRENGTH
RDENDBUDGET Government investment in PV Research Million US$ (2010
prices +exchange
rates)
POLINVEST Binary for presence of investment/tax Binary 0/1
policies
POLNETMET Binary for presence of net metering policies Binary 0/1
25
References
M.J. Bürer, R. Wüstenhagen, Which renewable energy policy is a venture capitalist’s best
friend? Empirical evidence from a survey of international cleantech investors, Energy
Policy 37 (12) (2009) 4997‐5006.
K. Cory, T. Couture, C. Kreycik, Feed‐in Tariff Policy: Design, Implementation, and RPS Policy
Interactions, National Renewable Energy Laboratory (NREL), Golden, Colorado, USA,
2009.
G.T. Gardner, P.C. Stern, Environmental Problems and Human Behavior, Allyn and Bacon,
Boston, 1996.
P. Gipe, Renewable Energy Policy Mechanisms, Mimeo, 2006.
G.M. Grossman, A.B. Krueger, Environmental Impacts of a North American Free Trade
Agreement, NBER Working Paper No. 3914, 1991. Also published in: P.M. Garber (Ed.),
The Mexico‐U.S. Free Trade Agreement, MIT Press, Cambridge, MA, 1993, pp. 13‐57.
R. Haas, G. Resch, C. Panzer, S. Busch, M. Ragwitz, A. Held, Efficiency and effectiveness of
promotion systems for electricity generation from renewable energy sources —
Lessons from EU countries, Energy 36 (4) (2011) 2186‐2193.
S. Jenner, Did Feed‐in Tariffs Work? An Econometric Assessment, Mimeo, Faculty of
Economics and Social Sciences, University of Tübingen, 2012.
S. Jenner, F. Groba, J. Indvik, Assessing the strength and effectiveness of renewable electricity
feed‐in tariffs in European Union countries, Energy Policy 52 (2013) 385‐401.
A.C. Marques, J.A. Fuinhas, J.R.P. Manso, Motivations driving renewable energy in European
countries: A panel data approach, Energy Policy 38 (11) (2010) 6877‐6885.
A.C. Marques, J.A. Fuinhas, Are public policies towards renewables successful? Evidence from
European countries, Renewable Energy 44 (2012) 109‐118.
M. Mendonça, Feed‐in Tariffs: Accelerating the Deployment of Renewable Energy, Earthscan,
London, 2007.
M. Mendonça, D. Jacobs, B. Sovacool, Powering the Green Economy: The Feed‐in Tariff
Handbook, Earthscan, London, 2009.
S. Owens, L. Driffill, How to change attitudes and behaviours in the context of energy, Energy
Policy 36 (12) (2008) 4412–4418.
T. Panayotou, Empirical Tests and Policy Analysis of Environmental Degradation at Different
Stages of Economic Development, Working paper 238, Technology and Employment
Programme, International Labor Office, Geneva, 1993.
G.R. Timilsina, L. Kurdgelashvili, P.A. Narbel, Solar energy: Markets, economics and policies,
Renewable and Sustainable Energy Reviews 16 (1) (2012) 449‐465.
W. White, A. Lunnan, E. Nybakk, B. Kulisic, The role of governments in renewable energy: The
importance of policy consistency, Biomass and Bioenergy 57 (2013) 1‐9.
26